Define agent purpose, success criteria, and evaluation scope into test artifacts.
npx skills add https://github.com/microsoft/eval-guide --skill eval-guide
Please help me install this Agent Skill. Skill: eval-guide Source: https://github.com/microsoft/eval-guide/tree/main/skills/eval-guide Command: npx skills add https://github.com/microsoft/eval-guide --skill eval-guide
This Skill provides a structured, end-to-end evaluation accelerator that helps teams articulate an agent's purpose, success criteria, and evaluation scope early, turning vague ideas into a concrete eval plan and test artifacts.
Describe your agent and run through Stage 0 to produce a ready-to-use eval plan.
High-intent search queries and answers about installing and using this skill.
Agent evaluation planning starts by defining the agent's purpose, success criteria, and evaluation scope. This Skill converts your agent vision into Stage 1 acceptance criteria, including a Value × Cost-of-Failure matrix and pass/fail conditions.
You can generate evidence-ready test artifacts by running your agent concept through the evaluation planner. It produces Stage 2 test cases in CSV format and Stage 1 documentation that import directly into Copilot Studio for review.
Stage-based evaluation is a structured framework that breaks agent assessment into Stage 0 through Stage 4 outputs. You need it when planning evals, generating test cases, interpreting results, and triaging failures for an agent concept or feature improvement.
Agent evaluation triage uses Stage 4 guidance to classify root causes when eval results arrive. It provides structured triage-ready outputs that help you interpret failures and identify why an agent concept or feature did not meet acceptance criteria.
Yes, the evaluation accelerator turns vague ideas into a concrete eval plan starting from Stage 0. You simply describe your agent concept and run through the discovery conversations to produce a ready-to-use plan without pre-existing test artifacts.
Yes, the evaluation scripts depend on the Anthropic SDK to function. You must have the @anthropic-ai/sdk installed in your environment to execute the stage-based evaluation and test case generation processes.
heximaoGenerate and optimize Python code for Claude API applications using the Anthropic SDK.
microsoftAnalyzes Copilot Studio evaluation CSV results to generate a structured triage report.
microsoftRegister Microsoft Entra ID apps and configure OAuth/MSAL workflows.
solace-illCoordinate qa-lead and qa-tester to generate test plans, cases, and sign-off reports.
